agent work verifier

The problem

Coding agents often report success on unfinished work. They leave behind pass statements, empty test assertions, and swallowed exceptions that result in false-positive green builds.

What it does

• Scans for structural stubs like NotImplementedError, ..., or docstring-only function bodies using Python AST.
• Detects "hollow" tests that contain no assertions or verifying calls.
• Identifies code that was defined but never referenced or wired up in the scanned tree.
• Locates forgotten debug statements like breakpoint(), debugger, or .only test flags.
• Catches swallowed exceptions and xfail markers that hide runtime failures.

Frameworks & tools

Works with Python 3.8+ (AST-based) and JavaScript/TypeScript (pattern-based). No external dependencies required.

Why this beats prompting it yourself

Generic LLM prompts often fail to catch their own omissions because the model "hallucinates" that the logic is there. This tool uses structural analysis to find objective evidence of unfinished work that a visual diff review will likely miss.

Use cases

• Pre-merge checks for AI-generated pull requests.
• Automated verification after an agent claims a task is complete.
• Identifying silent test failures in CI/CD pipelines.
• Cleaning up debug artifacts and dead code before shipping.

Known limitations

Python checks are structural and highly accurate, but JavaScript/TypeScript checks use pattern matching and may yield lower confidence results. It does not judge code quality, style, or performance.